Article | Published:

Latent analysis of unmodified biomolecules and their complexes in solution with attomole detection sensitivity

Nature Chemistry volume 7, pages 802809 (2015) | Download Citation

Abstract

The study of biomolecular interactions is central to an understanding of function, malfunction and therapeutic modulation of biological systems, yet often involves a compromise between sensitivity and accuracy. Many conventional analytical steps and the procedures required to facilitate sensitive detection, such as the incorporation of chemical labels, are prone to perturb the complexes under observation. Here we present a ‘latent’ analysis approach that uses chemical and microfluidic tools to reveal, through highly sensitive detection of a labelled system, the behaviour of the physiologically relevant unlabelled system. We implement this strategy in a native microfluidic diffusional sizing platform, allowing us to achieve detection sensitivity at the attomole level, determine the hydrodynamic radii of biomolecules that vary by over three orders of magnitude in molecular weight, and study heterogeneous mixtures. We illustrate these key advantages by characterizing a complex of an antibody domain in the solution phase and under physiologically relevant conditions.

Main

Protein–protein interactions are central to the framework with which biological systems respond to their environments over a range of temporal and spatial timescales1,2,3,4,5,6. Rapid and transient interactions in particular allow a system to react to a stimulus in a coordinated and complex fashion, using a limited number of components. Moreover, such interactions are increasingly recognized as a key target for the design of more selective pharmaceuticals that modulate, rather than broadly inactivate, their targets7. This approach—and indeed, the general aim of probing protein–protein interactions under physiologically relevant conditions—remains highly challenging, in part because many of the assays themselves have the potential to perturb the interactions under observation8,9.

Common methods for studying protein–protein interactions include two-hybrid screens10, mass spectrometry11, protein microarrays12 and surface plasmon resonance (SPR) techniques such as BIAcore. However, in exploiting fusion constructs, immobilization, or transfer into the gas phase, it can be challenging to relate the conformational ensemble sampled during the experiments to that explored under native conditions9. Methods such as analytical ultracentrifugation (AUC), isothermal titration calorimetry (ITC), nuclear magnetic resonance (NMR)13 and static and dynamic light scattering (DLS) allow operation under solution conditions, but often require concentrations exceeding the biologically relevant range and consume large amounts of protein or require specialized equipment14. Good detection limits have been attained with native fluorescence capillary electrophoresis (NCE)15 when the intrinsic fluorescence of the sample is high, and with backscattering interferometry (BSI)16 when factors other than binding events, which may cause a measurable change in refractive index, are precisely controlled.

Alternatively, a general and cost-effective method to increase the sensitivity is via the incorporation of protein or small molecule labels, such as fluorescent dyes, but even with careful choice of the labelling position these can introduce artefacts into the measurement by affecting the conformational ensemble of the protein studied17. Conventionally, if the protein is modified in some way—such as through the installation of a fluorescent protein or small molecule label—to permit sensitive detection, then the experiment is capable of revealing exclusively the behaviour of the modified protein, which may differ from that of the natural system. We present a new approach, which we define to be ‘latent’, because the labelling is instead a component of the measurement process such that it inherently does not perturb the biological process under observation (Fig. 1a). Latent labelling exploits the ability to spatially separate reaction chambers within a microfluidic system operating under steady-state flow conditions and continuously direct native, unmodified biomolecules with select properties to a chemically distinct region of the microfluidic system, where they are labelled with a fluorogenic molecule prior to the detection step (Fig. 1a,b). We thus exploit microfluidic and chemical tools to ensure that the behaviour of the unlabelled, physiologically relevant system can be extracted from the simple optical detection of the labelled system.

Figure 1: Latent labelling enables the development of native microfluidic analysis systems.
Figure 1

a, Conventional measurements can involve a compromise between sensitivity and accuracy—the behaviour of labelled biomolecules (with attached blue spheres) may differ from that of their unlabelled, physiological counterparts—but the introduction of a latent labelling step within the measurement process enables sensitive optical detection of labelled biomolecules to provide a snapshot of the time evolution of the native biomolecules. Under well-controlled laminar flow conditions in microfluidic systems, this time evolution can be related to fundamental physical properties such as charge18 or size (T. Müller et al., manuscript in preparation). b, To illustrate this, we have designed a native microfluidic diffusional sizing device. Arrows indicate the direction of fluid flow. Diffusion into a stream of buffer over time tD − t0 separates biomolecules according to their RH. Simulated distributions across the diffusion channel width before (t0, inset i) and after (tD, inset ii) this diffusion are shown for particles of two known RH. A well-defined fraction of the distribution (yellow rectangle) is selected and labelled by mixing with a fluorogenic molecule and denaturant (inset iii). Because labelling is quantitative (Fig. 2), optical detection of fluorescence intensity (inset iv) reveals the total concentration of biomolecules diverted for labelling, which reports the size distribution of the native, unlabelled species at time tD.

We demonstrate that biomolecules can be quantitatively labelled in seconds with a fluorogenic dye under denaturing conditions, on a microfluidic chip. We incorporate this latent labelling module into a native microfluidic diffusional sizing system. Using this technique, it is possible to detect even attomole quantities of biomolecules, while studying the behaviour of unlabelled biomolecules and their complexes under fully native solution conditions, by measuring and assessing changes in the hydrodynamic radius RH. We show that biomolecules varying in molecular weight (Mw) by over three orders of magnitude can be accurately sized without the need for a calibration step, and that this approach therefore exceeds the dynamic range of existing techniques. We further demonstrate that the technique tolerates both intrinsically disordered proteins and heterogeneous mixtures, and demonstrate the power of the method to characterize a clinically relevant α-synuclein immune complex. These results suggest that native microfluidic diffusional sizing, and indeed additional applications of the latent labelling approach, will become valuable tools for the characterization of biomolecules and their complexes, under fully native conditions and over a wide range of concentrations and timescales, and also for investigations into how such interactions can be modulated.

Results and discussion

Latent labelling and its integration into microfluidic systems

At a fundamental level, the measurement process consists of preparing a system in a well-defined initial state, monitoring its time evolution under a set of conditions governed by a parameter of interest and then detecting that time evolution through a change in some observable. If a label is installed to increase detection sensitivity, then the measurement reveals in such experiments exclusively the behaviour of the labelled system (Fig. 1a). We have used chemical and microfluidic tools to design a latent labelling strategy that decouples these steps of the measurement process, spatially and chemically, by confining them to distinct reaction chambers within a microfluidic system operating at steady state (Fig. 1a,b). Labelling enables highly sensitive detection, but the samples are unlabelled when the property of interest is probed via the time evolution step, so the label cannot affect the measured property.

Our approach exploits the distinct properties of fluids when confined to small (micrometre) length scales. When two streams of fluid, one containing biomolecules of interest and the other containing exclusively buffer, meet in a microfluidic channel (junction j1 in Fig. 1b), there is no convective mixing, and transport of the biomolecules into the buffer perpendicular to the direction of the flow lines proceeds, in the absence of an applied force, exclusively via diffusion19. It is possible to separate reaction chambers in time by spatially separating them along the direction of flow within the microfluidic system. Biomolecules of interest can be partitioned between these chambers by orienting the chambers perpendicular to the flow direction, such that only those biomolecules whose properties of interest fall within a desired range are able to flow along a particular path20,21.

Here we couple our latent labelling strategy with a diffusional sizing approach. Advantages of this approach include that sizing is done in the solution phase, is non-invasive, is suitable for the study of complex mixtures and is tolerant of a wide range of solution conditions. Biomolecular diffusion coefficients are determined by measuring the extent of mass transport perpendicular to the direction of flow, after a known residence time20,21,22,23,24. In the microfluidic device (Fig. 1b), the first step of the measurement process is the establishment of a well-defined initial state: when the protein and buffer streams meet at time t0, no diffusion has taken place, so proteins of all RH have the same initial distribution, laterally spanning half the width of the diffusion channel as the volumetric flow rates of the protein and buffer streams are equal. The system is then allowed to evolve for a defined period of time, tD − t0, after which the smaller species have diffused further than have the larger species. A simulation of the behaviour of two species, one with RH = 0.5 nm and the other with RH = 10 nm, is shown in Fig. 1b, insets i and ii. At this stage the biomolecule is unlabelled, so no direct observation of diffusion at time tD occurs. Instead, molecules that have diffused at least one-sixth of the channel width in time tD − t0 (yellow rectangle) are diverted to the labelling module. The total concentration of labelled molecules (integrating the size-dependent distribution within the yellow rectangle) reports the distribution at time tD, and is thus used to determine molecular size (see Fig. 3a, later).

We note the generality of the latent analysis approach. In the laminar regime, it is possible to choose a well-defined component of the flow and submit that component to the non-native conditions required for a detection platform of choice, with that detection delivering information about the entire distribution before the partition step and therefore on the behaviour of the unmodified sample under native conditions. Several detection methods, such as quartz crystal microbalance25 or nanospray mass spectrometry26 can be envisioned. Here, we have developed a fluorescence detection platform, because of its rapidity, high sensitivity, low cost and ease of implementation with standard laboratory instruments.

Defining the chemical requirements for latent labelling

The conceptual requirement that the total concentration of biomolecules be accessible from the measurement of fluorescence intensity alone defines the chemical requirements of the latent labelling module. To permit absolute concentration measurement on chip without the need for calibration, either each biomolecule should be labelled at a single site, or all potential reactive groups within the biomolecule should be modified. Owing to its generality and higher expected signal intensity, we explored the latter quantitative labelling approach, in which the biomolecular concentration can be determined whenever the number of reactive groups within the biomolecule is known. To facilitate quantitative modification and reduce quenching effects, all reactive groups within the biomolecule should be exposed to the solvent. The biomolecule should also be protected against precipitation if it passes through its isoelectric point (IEP) during the labelling step, and the entire process should reach completion on the second to minute timescale, to facilitate real-time readout. The use of a fluorogenic label, which becomes fluorescent exclusively upon reaction with groups of interest within the biomolecule, obviates the need for purification of the labelled biomolecules from the unreacted dye.

We addressed these challenges by adapting the fluorogenic reaction of ortho-phthalaldehyde (OPA) with primary amines27,28,29,30,31. Due to the abundance of primary amine moieties within protein molecules, this method is particularly suited for the determination of low protein concentrations, but any biomolecule containing at least one primary amine can be analysed. In the presence of compounds with thiol groups such as in β-mercapto ethanol (BME), reaction with a primary amine group—such as the protein N-terminal and lysine residues—forms a conjugated pyrrole ring, resulting in the formation of a substituted isoindole, giving rise to fluorescence in the blue region of the spectrum29,30. To ensure that all primary amines are exposed, our procedure involves the addition of high concentrations of sodium dodecyl sulfate (SDS) and excess reducing agent, conditions that are designed to denature the proteins at alkaline pH (10.5)32 (Fig. 2a).

Figure 2: Precise control of reaction time enables quantitative, fluorogenic protein labelling before detection.
Figure 2

a, Native proteins are denatured to expose all primary amine groups within the protein sequence for modification with non-fluorescent OPA. Labelling is quantitative if all or a constant proportion of primary amine groups are modified, such that regardless of sequence or structure, a single linear relationship describes the dependence of the generated isoindole fluorescence intensity on primary amine concentration for all of the peptides, proteins and free amino acids studied (see key). bd, Measurement of fluorescence intensity 3 s after mixing (b), just after the reaction has reached completion (c), permits accurate determination of approximately five orders of magnitude of protein concentration from the fluorescence intensity. Linearity is compromised 120 s after mixing (d) and at later time points (Supplementary Figs 2 and  3). e, The fluorescence intensity measured after 3 s on chip as a function of protein concentration down to 6 attomole, showing the viability of determining low protein concentrations on a microfluidic chip. Error bars show standard deviation among independent replicates. Data plotted with open markers have been excluded from the fit. Grey shaded areas indicate the range of concentrations over which bulk (b,d) and on-chip (e) absorption measurements are routinely possible (Supplementary Figs 4 and  5).

To assess the extent of reaction, we observed the fluorescence intensity generated upon labelling a set of well-characterized peptides and proteins with varying secondary, tertiary and quaternary structures, including some that would have passed through or approached their IEPs during the labelling step (Fig. 2, key). When the fluorescence intensity is plotted as a function of protein concentration, sequence-dependent relationships are observed as predicted. Supplementary Fig. 1 presents representative data for bovine serum albumin (BSA), lysozyme (Lys) and β-lactoglobulin (β-lac). If labelling is quantitative, however, such sequence-dependent variation should disappear when fluorescence intensity is instead plotted as a function of the concentration of reactive groups, that is, primary amines. The fluorescence intensity of each protein within the reference set should then fall on a single line, passing through the origin, that also describes the fluorescence intensity of free glycine and lysine. These amino acid controls are small molecules that are chemically similar to the labelled groups, but which are solvent-accessible without denaturation.

Because the isoindole fluorophore formed during the labelling reaction lacks chemical stability33,34,35, extraction of quantitative information from this labelling technique under microfluidic conditions requires that the fluorophore is continuously generated under flow and measured at a defined time after mixing. We examined the dependence of isoindole fluorescence intensity on primary amine concentration when our reference set of proteins and amino acids was measured 3 s after mixing (Fig. 2b), the time we had determined was required for the reaction to reach completion on a microfluidic chip (Fig. 2c). Indeed, a single linear relationship describes the dependence of fluorescence intensity on primary amine concentration over a range of more than three orders of magnitude of primary amine concentration, which corresponds to nearly five orders of magnitude in protein concentration. The fit is fluorescence = 1.11 × (primary amine concentration) with R2 = 0.91 (see Supplementary Methods), and a variation in labelling efficiency with protein sequence or among the amino acid controls is not observed.

Precise control over the reaction time, as is accessible via a microfluidic platform, is crucial. When the fluorescence intensity was instead measured at the 120 s time point, we observed deviations from linearity at high and low protein concentrations, as well as substrate-dependent variations in labelling efficiency (Fig. 2d). The fit is fluorescence = 1.16 × (primary amine concentration), R2 = 0.72. We investigated the reason for these deviations and found that although fluorescence is rapidly generated upon mixing, at later time points the fluorescence intensity generally decreases in a complex substrate- and concentration-dependent manner (Supplementary Figs 2 and 3), consistent with the reported differences in the degradation rates for isoindole fluorophores formed via reaction with varying substrates34,35. Measurement at the optimized 3 s time point (used for the remaining experiments in this Article) allowed protein concentrations as low as 1 nM to be determined (Fig. 2b). This corresponds to a concentration approximately 2,000 times less than the concentration of protein that can be routinely measured via absorption (Supplementary Fig. 4). We note that if the study of protein concentrations exceeding the linearity limit is desired, the volumetric mixing ratio of the dye and protein streams can be altered, or a rapid on-chip dilution module can be incorporated36.

Given the low path lengths characteristic of microfluidic systems, we also explored the detection limit accessible on the diffusional sizing device (Fig. 1b), using BSA as a test system. We measured the fluorescence signal on chip for between 3.75 nM and 15 µM of this protein, as shown in Fig. 2e, suggesting a detection limit in the low nM range. Given the 1.6 nl volume of the portion of the detection region in which fluorescence intensity is quantified in the microfluidic device (Fig. 1b and Supplementary Fig. 7), this is equivalent to the quantification of on the order of 10−17 mol (10 attomole) protein on our chips.

Accurate protein sizing with high dynamic range

Molecular size is calculated based on simple measurements of fluorescence intensity. Quantitative labelling ensures that absolute protein concentration can be determined from the measurement of fluorescence intensity within the detection region. Because the system is at steady state, measurement of the protein concentration downstream of the latent labelling module reveals the total concentration of the protein diverted for labelling at time tD (Figs 3a and  1b, yellow rectangle) to the junction j2 (Fig. 1b). Because the microfluidic system is time-independent, it is possible to increase the integration time so as to permit detection of very low concentrations of biomolecules.

Figure 3: Sizing proteins, heterogeneous mixtures and protein complexes.
Figure 3

a, Concentration profiles of reference species of known size (T. Müller et al., manuscript in preparation) resulting from diffusion and advection (ref. 20) calculated for the channel geometry used, viewed perpendicularly to the flow direction across the diffusion channel in Fig. 1b. A reference distribution of species distributed homogeneously across the channel (black solid line) and the initial distribution before diffusion (black dashed line) have the same profiles for species of all sizes, and after diffusion these are determined by the RH of the species, indicated with the colour scale. The RH is thus obtained by comparing the total concentrations of diffused versus homogeneously distributed species selected for labelling (yellow shaded area), as shown in the inset (Int., intensity). The homogeneous distribution was achieved by flowing the analyte into the device from both the 'biomolecule' and 'buffer' channels in Fig. 1b. b, RH values determined in this manner are compared to those obtained with PFG-NMR (lilac) and AUC (aqua) (Supplementary Table 1). The studied molecules vary by over three orders of magnitude in molecular weight and include intrinsically disordered proteins and heterogeneous mixtures. c, We exploit this heterogeneous mixture tolerance and accurate sizing of disordered, as well as folded, structures to characterize a novel Parkinson's-related immune complex between a nanobody and equimolar (5 µM) α-synuclein, by characterizing the hydrodynamic radii of all components. Our data suggest that the nanobody interacts stoichiometrically with α-synuclein monomers. The schematic is based on the known locations of a related nanobody's interaction38 with an ensemble representing intrinsically disordered conformations of α-synuclein39,40,41. Throughout, error bars represent standard deviation among independent replicates.

To improve the robustness of the fluorescence intensity quantification against well-known effects such as a variation in the illumination source intensity over time37, a second measurement was taken of a homogeneous reference distribution. Practically, this calibratory measurement can be made most readily by simply loading the same analyte solution into both the ‘biomolecule’ and ‘buffer’ inlets shown in Fig. 1b. Species of varying RH (indicated colorimetrically) differ in the fraction of diffusing species that are diverted for labelling compared with homogeneously distributed (Fig. 3a, black line) species. This is assessed experimentally by comparing the fluorescence intensities for these species in the detection region of the device (Fig. 3a, inset). Given the total volume of protein flowing through the device during the measurement (diffusion + labelling + integration time), both for the sample and the homogeneously distributed reference molecules, on the order of one femtomole of protein is required for a sizing measurement, although miniaturization of the device could reduce this requirement even further.

To assess the accuracy of the RH values obtained with our system, we designed a ‘sizing ladder’ of biomolecules varying by over three orders of magnitude in Mw. In order of increasing RH, these included lysine, a heterogeneous mixture of insulin monomer and dimer, β-lac dimer, intrinsically disordered α-synuclein, BSA, a covalent BSA dimer and β-galactosidase tetramer. This set of molecules includes proteins that differ in secondary and tertiary structure, are natively unfolded as well as folded, and exist as monomeric species or as complexes. In Fig. 3b, we compare the results we obtained to those reported in the literature using two established methods for measuring the RH values of unlabelled proteins (see Supplementary Table 1 for literature references). Pulsed-field gradient NMR (PFG-NMR) was used for low-Mw species with low extinction coefficients, and analytical ultracentrifugation (AUC) was used for higher-Mw species. Both values were reported where possible. We found that the RH values obtained with microfluidics agreed closely with those obtained from the composite of AUC and PFG-NMR techniques over the entire Mw range studied. The RH values obtained with microfluidics were also consistently more accurate than Mw-based predictions of molecular size (Supplementary Fig. 8). The accuracy of our method was particularly evident in our analysis of α-synuclein, an intrinsically disordered protein implicated in Parkinson's disease. We found that the microfluidic RH value obtained for α-synuclein was consistent with that obtained via PFG-NMR42 and that both were larger than that obtained with AUC43, a result that is expected because the natively unfolded structure of α-synuclein is not compact and would be expected to sediment more slowly.

Characterizing native protein–protein interactions

Having established the sensitivity and accuracy of the system, we also explored its tolerance of heterogeneous mixtures. The bovine insulin hormone has been studied extensively in vitro. At low pH and in the absence of Zn2+, the monomer and dimer predominate. Its well-defined oligomers have been characterized hydrodynamically and thermodynamically, so we chose this system to explore whether the composite hydrodynamic radii we obtained reflect the radii expected based on the relative abundance of each species present in the mixture. At pH 2 and in the absence of Zn2+, an association constant of 1.1 × 104 M−1 has been reported44, which should under the conditions of this study result in 71% insulin monomer and 29% insulin dimer. The RH of 1.64 nm obtained with native microfluidic diffusional sizing reflects the proportions of monomer and dimer—which have RH values of 1.60 (ref. 45) and 1.78 nm (ref. 46), respectively—present in the sample.

Finally, we made use of this heterogeneous mixture tolerance to explore the use of this procedure—as a proof of concept— to characterize an undescribed protein–protein interaction. We chose to study an interaction between α-synuclein and a single-domain camelid antibody, termed a nanobody47. Due to their small size and high stability and specificity, nanobodies are rapidly emerging as important research tools in structural biology and medicine, including for particular use as diagnostic markers and therapeutics for protein misfolding diseases48,49. We have developed nanobodies to study the misfolding of several proteins and have shown that the nanobody NbSyn238 can be used as a molecular probe for the detection of subtle conformational differences upon α-synuclein fibril maturation50. Due to the conformational flexibility of intrinsically disordered α-synuclein, the characteristics of the labelled complex may be expected to differ significantly from those of the native, unmodified system.

We used the microfluidic platform to explore the binding of α-synuclein to a variant of NbSyn2, NbSyn138. Our approach involved characterizing the hydrodynamic radii of all components (Fig. 3c). If NbSyn1 binds to the α-synuclein monomer, under conditions such as those explored here in which the system would be expected to be fully bound based on the dissociation constants of related mutants, then the predicted size increase on binding can be calculated based on the change in molecular weight on complex formation, using a Mw-based prediction of molecular size (as described in Supplementary Fig. 8). Our prediction corrects for the larger than ‘minimum’ RH observed for α-synuclein in isolation due to its intrinsically disordered structure by taking into account an adjusted molecular weight that would be expected for a protein with the measured RH. In this manner, the size of the bound complex is predicted to be 3.11 nm, which is in accord with the measured 3.10 ± 0.35 nm and suggests that NbSyn1 indeed binds to the α-synuclein monomer. By contrast, binding to an oligomeric species would require a complex size of at least 3.82 nm, which is not observed. Thus, this approach reveals not only that a binding event has occurred, but also suggests the oligomerization state of the target, which is frequently, as in the Parkinson's-associated system studied, a crucial determinant of its biological activity6.

Native microfluidic diffusional sizing has a combination of features that are distinct from those of conventional methods (such as SPR and ITC) for characterizing this type of complex: measurements are rapid, entirely in the solution phase, tolerant of any desired buffer conditions as long as the only primary amines are those intended for detection, consume only microlitres of sample, can probe interactions that are entropically as well as enthalpically driven, and the direct output of the measurement is a fundamental physical property—molecular size—which gives additional information about the oligomerization state of the target.

Conclusions

In summary, the development of methods enabling the characterization of native, unmodified biomolecules and their complexes in the solution phase is of central significance in structural and functional biology. The data presented here demonstrate that, by integrating chemical and physical tools to create a latent analysis approach, our technology makes it possible to achieve highly sensitive (attomole) detection sensitivity of native proteins. Species ranging in size from individual amino acids (146 Da) to large protein complexes (464,000 Da) could be accurately sized. This method is applicable to intrinsically disordered proteins and heterogeneous mixtures and has key advantages over existing technologies. Essentially, it is a non-disruptive, solution-phase method that enables characterization of low concentrations of biomolecules and their interactions, without the need for prior labelling. This is important, because if prior labelling is used to enhance detection it can both perturb the behaviour under observation and require significant researcher effort to identify the conditions that may lessen this perturbation. Indeed, due to its simplicity and generality, we expect that this technology, and other latent analysis approaches in which the behaviour of unmodified biomolecules is measured with high sensitivity, will be of particular relevance in the study of the strength and kinetics of protein–protein and protein–nucleic acid interactions (which are increasingly recognized as a new generation of pharmaceutical targets8) as well as in the characterization of protein and protein complex interactions with small-molecule modulators.

Methods

Bulk labelling measurements

A variety of fluorogens, stoichiometries and denaturing conditions were investigated using a fluorescence spectrometer (Varian, Cary Eclipse) and quartz fluorescence cuvettes (Hellma) or a fluorescence microplate reader (BMG LabTech) and half-area non-protein binding microplates (Corning, product #3881). The quantitative labelling cocktail described in this Article was composed of 12 mM OPA, 18 mM BME and 4% wt/vol SDS in 200 mM carbonate buffer, pH 10.5. Generally, 16 mg OPA was dissolved in 4 ml of 500 mM carbonate buffer, pH 10.5. Then, 12.63 µl BME was added, together with 4 ml water and 2 ml of a 20% wt/vol solution of SDS. The cocktail was protected from light and heated at 65 °C for 10 min, or until the OPA dissolved, then allowed to cool to room temperature and filtered. Labelling solutions were protected from light at room temperature and used within five days of preparation, or frozen and used within fourteen days of preparation. The solutions were standardly mixed at 1:1 vol/vol with each of the samples of interest. Unless otherwise stated, protein solutions were prepared in 5 mM HEPES, pH 7.0, and their concentrations were determined spectrophometrically on a NanoDrop UV–vis spectrophotometer (Thermo Scientific).

Time-controlled fluorescence measurements were performed using a CLARIOstar microplate reader (BMG LabTech) fitted with an injector module. The measurements were performed in ‘well mode’, meaning that each well was treated separately. The injector module delivered 50 µl dye into a single well at a speed of 430 µl s–1, agitated the plate for 1 s, and then measured the sample every 0.25 s for a duration of 125 s, before moving on to the next well. Every sample and dye background solution was prepared in triplicate.

For information on the biomolecules used in this study and for a discussion of microfluidic design and fabrication, see Supplementary Sections 1 and 3, respectively.

Microfluidic measurements

Devices were loaded by first filling all channels from the outlet with the appropriate buffer. The buffer and samples were filtered through a 0.22 µm filter (Millipore) immediately before use, to eliminate particulate matter that could clog the devices. Generally, either a 1 ml Hamilton glass syringe or a 1 ml plastic Air-Tite syringe was used, connected through a 27-gauge needle to portex tubing. No differences were noted between the performance of glass and plastic syringes at the flow rates used in these experiments. Pressure was then applied simultaneously at the inlets and through the syringe to remove any bubbles formed during the loading process, and reagents were introduced with gel loading tips at the device inlets. Reagent loading varied between 10 and 200 µl, depending on the nature of the particular experiment, although even smaller volumes can be used.

Fluid was withdrawn through the device with a neMESYS syringe pump. To draw reagents through the device initially and minimize the effects of any inlet cross-flow during the loading step, 20 µl fluid was first withdrawn at a flow rate of 300 µl h–1. For the microfluidic device used in these experiments, a 25 µl h–1 flow rate in the diffusion chamber was selected, corresponding to a 33.3 µl h–1 withdrawal rate at the outlet. The flow rate was allowed to equilibrate for at least 18 min before the start of image acquisition and for at least 500 s following sample changes. The initial equilibration steps can be performed with buffer to reduce sample consumption. An efficient sample change procedure involved depositing a drop of buffer around the gel-loading tip that delivered the sample into the indicated inlet, then rapidly removing that tip and replacing it drop to drop with a second tip filled with the new sample, all the while withdrawing at the desired flow rate.

Bright-field and fluorescence images were acquired using a Zeiss AxioObserver Microscope, fitted with an Evolve 512 charge-coupled device camera (Photometrics) and a 365 nm Cairn OptoLED and DAPI filter (product # 49000, Chroma) for the fluorescence images. A series of objectives (×2.5, ×5, ×10 and ×20) were used, and exposure times of between 10 ms and 10 s were applied. Generally, between 10 and 60 images were averaged during each acquisition. When the signal intensity was low, electron-multiplying (EM) gain was used, or adjacent pixels were binned. For each set of measurements, at least one dye background image was taken to account for the (minimal) fluorescence of the unreacted dye. A flatfield background image was also acquired and measurements were taken in a dark environment, with the temperature maintained at 25 °C.

References

  1. 1.

    , , & Adapting proteostasis for disease intervention. Science 319, 916–919 (2008).

  2. 2.

    et al. Stilbene vinyl sulfonamides as fluorogenic sensors of and traceless covalent kinetic stabilizers of transthyretin that prevent amyloidogenesis. J. Am. Chem. Soc. 135, 17869–17880 (2013).

  3. 3.

    & Prions: protein aggregation and infectious diseases. Phys. Rev. 89, 1105–1152 (2009).

  4. 4.

    et al. The presence of an air–water interface affects formation and elongation of α-synuclein fibrils. J. Am. Chem. Soc. 136, 2866–2875 (2014).

  5. 5.

    et al. Understanding amyloid aggregation by statistical analysis of atomic force microscopy images. Nature Nanotech. 5, 423–428 (2010).

  6. 6.

    & Protein misfolding, functional amyloid, and human disease. Annu. Rev. Biochem. 75, 333–366 (2006).

  7. 7.

    & The road less traveled: modulating signal transduction enzymes by inhibiting their protein–protein interactions. Curr. Opin. Chem. Biol. 13, 284–290 (2009).

  8. 8.

    & Reaching for high-hanging fruit in drug discovery at protein–protein interfaces. Nature 450, 1001–1009 (2007).

  9. 9.

    , & Methods for the detection and analysis of protein–protein interactions. Proteomics 7, 2833–2842 (2007).

  10. 10.

    et al. A comprehensive analysis of protein–protein interactions in Saccharomyces cerevisiae. Nature 403, 623–627 (2000).

  11. 11.

    et al. Systematic identification of protein complexes in Saccharomyces cerevisiae by mass spectrometry. Nature 415, 180–183 (2002).

  12. 12.

    , & Label-free detection methods for protein microarrays. Proteomics 6, 5493–5503 (2006).

  13. 13.

    Mapping protein–protein interactions in solution by NMR spectroscopy. Biochemistry 41, 1–7 (2001).

  14. 14.

    , & Free-solution, label-free protein–protein interactions characterized by dynamic light scattering. Biophys. J. 98, 297–304 (2010).

  15. 15.

    & High-sensitivity laser-induced fluorescence detection of native proteins in capillary electrophoresis. J. Chromatogr. A 595, 319–325 (1992).

  16. 16.

    et al. Free-solution, label-free molecular interactions studied by back-scattering interferometry. Science 317, 1732–1736 (2007).

  17. 17.

    , , & Preparation of fluorescently-labeled amyloid-beta peptide assemblies: the effect of fluorophore conjugation on structure and function. J. Mol. Recogn. 22, 403–413 (2009).

  18. 18.

    et al. Integration and characterization of solid wall electrodes in microfluidic devices fabricated in a single photolithography step. Appl. Phys. Lett. 102, 184102–4 (2013).

  19. 19.

    Theoretical Microfluidics (Oxford Master Series in Physics, Oxford Univ. Press, 2008).

  20. 20.

    & Diffusion-based extraction in a microfabricated device. Sens. Actuat. A 58, 13–18 (1997).

  21. 21.

    & Microfluidic diffusion-based separation and detection. Science 283, 346–347 (1999).

  22. 22.

    , , & Quantitative analysis of molecular interaction in a microfluidic channel: the T-sensor. Anal. Chem. 71, 5340–5347 (1999).

  23. 23.

    et al. A rapid diffusion immunoassay in a T-sensor. Nature Biotechol. 19, 461–465 (2001).

  24. 24.

    , & Optical measurement of transverse molecular diffusion in a microchannel. Biophys. J. 80, 1967–1972 (2001).

  25. 25.

    , , & in Methods in Molecular Biology Vol. 752 (eds Hill, A. F., Barnham, K. J., Bottomley, S. P. & Cappai, R.) 137–145 (Humana, 2011).

  26. 26.

    , , & Recent advances in microfluidics combined with mass spectrometry: technologies and applications. Lab Chip 13, 3309–3322 (2013).

  27. 27.

    Fluorescence reaction for amino acids. Anal. Chem. 43, 880–882 (1971).

  28. 28.

    & o-Phthalaldehyde fluorogenic detection of primary amines in the picomole range. comparison with fluorescamine and ninhydrin. Proc. Natl Acad. Sci. USA 72, 619–622 (1975).

  29. 29.

    & The structure of the fluorescent adduct formed in the reaction of o-phthalaldehyde and thiols with amines. J, Am. Chem. Soc. 98, 7098–7099 (1976).

  30. 30.

    , & Rational design and evaluation of improved o-phthalaldehyde-like fluorogenic reagents. Anal. Biochem. 144, 233–246 (1985).

  31. 31.

    , , , & Microchip capillary electrophoresis with an integrated postcolumn reactor. Anal. Chem. 66, 3472–3476 (1994).

  32. 32.

    Protein–surfactant interactions: a tale of many states. Biochim. Biophys. Acta 1814, 562–591 (2011).

  33. 33.

    , , & Factors affecting the stability of fluorescent isoindoles derived from reaction of o-phthalaldehyde and hydroxyalkylthiols with primary amines. Anal. Biochem. 135, 495–504 (1983).

  34. 34.

    & High performance liquid chromatographic determination of subpicomole amounts of amino acids by precolumn fluorescence derivatization with o-phthaldialdehyde. Anal. Chem. 51, 1667–1674 (1979).

  35. 35.

    , , & The interaction of amino acids with o-phthalaldehyde: a kinetic study and spectrophometric assay of the reaction product. Anal. Biochem. 101, 188–195 (1980).

  36. 36.

    et al. Single-molecule measurements of transient biomolecular complexes through microfluidic dilution. Anal. Chem. 85, 6855–6859 (2013).

  37. 37.

    Accuracy and precision in quantitative fluorescence microscopy. J. Cell Biol. 185, 1135–1148 (2009).

  38. 38.

    et al. Structure and properties of a complex of alpha-synuclein and a single-domain camelid antibody. J. Mol. Biol. 402, 326–343 (2010).

  39. 39.

    , , & Statistical coil model of the unfolded state: resolving the reconciliation problem. Proc. Natl Acad. Sci. USA, 102, 13099–13104 (2005).

  40. 40.

    , , & Structure and dynamics of micelle-bound human α-synuclein. J. Biol. Chem. 280, 9595–9603 (2005).

  41. 41.

    , , , & Mapping long-range interactions in α-Synuclein using spin-label NMR and ensemble molecular dynamics simulations. J. Am. Chem. Soc. 127, 476–477 (2005).

  42. 42.

    , , & Solvent-induced collapse of α-synuclein and acid-denatured cytochrome c. Prot. Sci. 10, 2195–2199 (2001).

  43. 43.

    , , , & NACP, a protein implicated in Alzheimer's disease and learning, is natively unfolded. Biochemistry 35, 13709–13715 (1996).

  44. 44.

    , & Insulin self-association. Spectrum changes and thermodynamics. Biochemistry 12, 4385–4392 (1973).

  45. 45.

    et al. Structure of human insulin monomer in water/acetonitrile solution. J. Biomol. NMR 40, 55–64 (2008).

  46. 46.

    & Detection of insulin aggregates with pulsed-field gradient nuclear magnetic resonance spectroscopy. Anal. Biochem. 229, 214–220 (1995).

  47. 47.

    Nanobodies: natural single-domain antibodies. Annu. Rev. Biochem. 82, 775–797 (2013).

  48. 48.

    , & Antibodies and protein misfolding: from structural research tools to therapeutic strategies. Biochim. Biophys. Acta 1844, 1907–1919 (2014).

  49. 49.

    & in Methods in Molecular Biology Vol. 911 (eds Saerens, D. & Mulydermans, S.), 533–558 (2012).

  50. 50.

    et al. Nanobodies raised against monomeric α-synuclein distinguish between fibrils at different maturation stages. J. Mol. Biol. 425, 2397–2411 (2013).

Download references

Acknowledgements

The authors acknowledge the European Research Council, Biotechnology and Biological Sciences Research Council, Wellcome Trust, Newman Foundation, Winston Churchill Foundation and Elan Pharmaceuticals for financial support. E.D.G. was supported by the Medical Research Council (G1002272). The authors thank J. Steyaert at the Free University of Brussels for sharing the NbSyn1a clone.

Author information

Affiliations

  1. Department of Chemistry, University of Cambridge, Lensfield Road, Cambridge CB2 1EW, UK

    • Emma V. Yates
    • , Thomas Müller
    • , Luke Rajah
    • , Erwin J. De Genst
    • , Paolo Arosio
    • , Michele Vendruscolo
    • , Christopher M. Dobson
    •  & Tuomas P. J. Knowles
  2. Department of Biochemistry and Structural Biology, Lund University, Lund SE221 00, Sweden

    • Sara Linse

Authors

  1. Search for Emma V. Yates in:

  2. Search for Thomas Müller in:

  3. Search for Luke Rajah in:

  4. Search for Erwin J. De Genst in:

  5. Search for Paolo Arosio in:

  6. Search for Sara Linse in:

  7. Search for Michele Vendruscolo in:

  8. Search for Christopher M. Dobson in:

  9. Search for Tuomas P. J. Knowles in:

Contributions

T.P.J.K. and C.M.D. supervised the research. E.V.Y., L.R., M.V., C.M.D. and T.P.J.K. conceived and designed the experiments. E.V.Y. performed the experiments. E.V.Y. and T.M. analysed the data. E.J.D.G., P.A. and S.L. contributed materials and/or analysis tools. E.V.Y., C.M.D. and T.P.J.K. wrote the paper, and all authors commented on the paper.

Competing interests

Part of the work described here has been the subject of a patent application filed by Cambridge Enterprise Ltd, a fully owned subsidiary of the University of Cambridge (now licensed to Fluidic Analytics, of which C.M.D. is a scientific advisor and T.P.J.K. is a board member).

Corresponding authors

Correspondence to Christopher M. Dobson or Tuomas P. J. Knowles.

Supplementary information

PDF files

  1. 1.

    Supplementary information

    Supplementary information

About this article

Publication history

Received

Accepted

Published

DOI

https://doi.org/10.1038/nchem.2344

Further reading